insightslisted

# Insights You produce a **decision-grade** report on how the user works with their coding agents — not a wall of charts they nod at and forget. The original `weekly-insights` told people "your sessions are long" and stopped. That is the failure mode this skill exists to kill. Two principles drive everything: 1. **Action-first.** Every observation — including *what's working* — must end in a concrete action with a number attached. A bare note is a bug. See `references/actionability-contract.md`. 2. **Multi-provider, honest about depth.** You report across Claude, Codex, and agy/Gemini. Claude has rich facets (goals/friction/outcomes); Codex and agy do **not** — their insights are quantitative/behavioral only, and you must say so rather than fabricate narrative for them. > Reference files sit alongside this SKILL.md — load them as you work: > - `references/actionability-contract.md` — the Finding→Evidence/threshold→Action→Effect shape, the default thresholds table, the "level-up the good" rule, and vague-vs-actionable examples. **Load before writing any narrative.** > - `references/html-template.md` — the extended HTML: provider switcher, Trend section, Action Plan, plus the original CSS/charts. **Load before rendering.** > - `references/data-and-history.md` — provider transcript locations, what signal each yields, the history-dir layout, the three-artifact-per-period scheme, how trend comparison reads prior runs cheaply, **and the Otto-generated-facets cache (